The following content has been provided by the University of Erlangen-Nürnberg.

So good morning everybody. First I have to announce that tomorrow we will have no lecture.

I will be in Oslo tomorrow and I think it's just 45 minutes so we should skip tomorrow

and we have the lecture today. I'll speed up a little bit so I will cover all the topics that I wanted to do tomorrow as well.

And next week we will continue on Monday. Okay? Sounds like a good deal.

Before we continue in the text, let me briefly summarize what we have considered so far.

So we talk about interventional medical image processing and the interventional medical image processing

basically deals with the problem of building systems, image processing devices that are used during the patient's treatment.

So the doctor is next to the patient and treats, for instance, here an aneurysm.

And tries to fix problems with the support of the image information that we provide.

And we have considered so far various concepts that are important to do on the fly image processing.

For instance, we have discussed a few approaches to feature detection.

I have introduced to you gradients. I introduced to you the structure tensor

that is easily explained by looking at gradients in the local neighborhood.

We talked about SIFT features. We talked about HOC features.

We also have seen pre-processing methods like the bilateral filter.

And we have seen two weeks ago also that GPU implementations are becoming more and more important in this environment.

Why? Because GPUs provide computational power like dedicated hardware devices 10 years ago.

And today with GPUs we basically can achieve any performance that is required for an interventional treatment.

Basically. There are a few exceptions.

But many things can be implemented very efficiently on GPUs using CUDA or OpenCLS programming languages.

As you might have heard of.

So that's that. And then we talked about shutter segmentation.

So how to find exactly the part of the human body that we want to exhibit for radiation.

And in this context we also learned something about half transform.

And there should also be a chapter of a book in the web where the half transform is explained in detail again

so that you can follow up with all these ideas.

The half transform is good for what?

For straight line detection.

It's used for straight line detection and basically you need to know how to compute the gradient or the slope of a straight line

and how to represent straight lines basically in terms of math.

And then you can easily understand the core ideas of the half transform.

And then we started with a very practical chapter on magnetic navigation.

Where we basically considered the problem we have here on aneurysm

and we want to bring the catheter into the aneurysm without using mechanical control.

The mechanical control of the catheter like it is done in most hospitals today.

So we will apply basically here magnetic forces from the outside

that basically direct the catheter into the direction where we want to have it.

And one long term vision here is that we have a CT, we do all the planning

and basically the magnetic field is automatically computed on the fly

while we are just pushing the catheter into the human brain for instance.

Today it's still different. We are not yet at this point.

Today doctors are required to adjust the orientation of the magnetic field step by step.

And it's also quite interesting.

We are currently building a system with Siemens that does the following.

This is the stomach.

And we build a system for capsule endoscopy.

And what we do is basically here you have a capsule with a little camera in it.

And the patients they have to drink water.

So they have a water belly.